hbseong/HarmAug_generated_dataset
收藏Hugging Face2025-03-05 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/hbseong/HarmAug_generated_dataset
下载链接
链接失效反馈官方服务：
资源简介：
HarmAug数据集包含了使用HarmAug方法生成的提示语（prompt）和响应（response），用于安全守卫模型的知识蒸馏的有效数据增强。该数据集同样用于训练HarmAug守卫模型。数据集中的unsafe-score由Llama-Guard-3测量，对于没有响应的行，unsafe-score表示提示语的不安全性；对于有响应的行，unsafe-score表示响应的不安全性。
This dataset contains generated prompts and responses using HarmAug: Effective Data Augmentation for Knowledge Distillation of Safety Guard Models. It is also used for training our HarmAug Guard Model. The unsafe-score is measured by Llama-Guard-3. For rows without responses, the unsafe-score indicates the unsafeness of the prompt. For rows with responses, the unsafe-score indicates the unsafeness of the response.
提供机构：
hbseong
